The question lands in every retainer kickoff. We can earn the citations. What is one of them worth.

The honest version of the answer is “we can model it but with caveats.” The dishonest version is a confident dollar figure with no methodology behind it. This piece is the modelling we use, with the caveats kept visible so a CFO can still trust the number.

For B2B SaaS our portfolio median lands between $850 and $1,400 per active citation per quarter. Crypto and fintech run 2 to 3 times higher because deal sizes are. Local services run 5 to 10 times lower because citation volume is generous but session value is small. The methodology that gets you to those numbers is below.

Why this is hard

AI citations carry no UTM. Perplexity passes a referer occasionally, ChatGPT and Claude almost never, Gemini’s pass-through is unreliable, Google AI Overview drives clicks that mostly look like direct traffic. So the analytics platform sees the visit as “direct” or “organic - branded query” and the citation that earned it leaves no fingerprint.

That is the bad news. The good news is that the lift is real and large enough to detect with the right comparison. Three independent signals carry the value — and once you measure all three, the dollar number falls out.

The three lifts AI citations actually create

Branded-search uplift. Someone reads about your brand inside an AI answer. They do not click the citation. Two days later they Google your brand name to read more. The session lands as branded organic search. Your conversion rate on branded search is your normal conversion rate on branded search — the visit is high-intent, high-value.

Assisted-conversion direct sessions. Someone reads about you inside an AI answer. They click the citation, browse, do not convert, leave. Three days later they come back direct (typed the URL, bookmark, or remembered the name). They convert. The first session was AI, the conversion gets attributed to direct.

Forwarded recommendations. A buyer asks ChatGPT “best AEO agency for SaaS.” Three names come back. The buyer pastes the list into a Slack thread or forwards it to a colleague. Another person clicks the name a week later. There is no fingerprint at all on this kind of lift but it is the most valuable single thing AI search does for B2B brands.

You cannot measure forwarded recommendations directly. You can model them as a multiplier on the first two — typically 1.3 to 1.6 in B2B contexts based on what we have triangulated from client interviews.

Method 1 — branded-search uplift

Easiest single proxy. Defensible to anyone who looks at the maths.

You need a control period and an exposure period. Pick a prompt cluster where you know you started being cited on a specific date — say March 14, when our tracker first logged a Perplexity citation on “best AEO agency for SaaS.” The control period is the 28 days before that date. The exposure period is the 28 days after.

In Google Search Console, isolate the branded queries (the brand name with various qualifiers — “answerly agency,” “answerly agency reviews,” “answerly aeo,” etc.). Sum the impressions for the control window. Sum them for the exposure window. The difference is the lift.

Apply your branded-search CTR — usually around 35% for top-of-page positions — to get visits. Multiply by your branded-search conversion rate. Multiply by your average deal value. Divide by the number of distinct active citations in the exposure window.

The number you get is the branded-search component of citation value. In our portfolio this lands between $300 and $600 per citation per quarter for B2B SaaS.

Method 2 — assisted conversions on direct sessions

This one needs an analytics setup that tracks first-touch and last-touch separately. GA4 with cross-channel attribution works once you accept the model’s limitations.

Run a 14-day attribution window. For B2B SaaS that captures roughly 70% of the path-to-conversion length we see in our retainer data. Ecom is 7 days, crypto/fintech 30, local services 24 hours.

Tag a “direct” segment that excludes navigational direct (people who already know your brand — typed the URL because they were referred). The cleanest proxy is direct sessions with no prior brand interaction in the analytics history. Imperfect but defensible.

Count the conversions in that segment over the exposure window. Subtract the same count over the control window — same length, immediately prior. The difference is the assisted-conversion lift attributable to AI exposure.

Divide by active citation count. In our portfolio this lands at $400 to $700 per citation per quarter for B2B SaaS — close to the branded-search number, sometimes higher.

Method 3 — half-life adjustment

This is the line CFOs notice. A citation that lasts 21 days is not the same asset as a citation that lasts 90 days. Without the half-life adjustment your model overcounts.

The simplest correction — multiply each citation’s dollar value by (actual lifespan / target horizon). If you are modelling on a quarterly horizon (90 days), a citation that lasts 18 days contributes (18/90 = 0.20) of its potential value. A 90-day citation contributes 1.0.

The half-life numbers come from your own tracking, ideally, or from our citation half-life study as a starting estimate. Our portfolio medians by content type — news 10 days, tactics 30–60, evergreen 90–180.

Once you apply the adjustment, the headline number drops by 30 to 50% for most clients. That is the honest one. The 30 to 50% you are not counting is the citations that expired before they could create their full value.

A worked example

Pretend you are a B2B SaaS doing $30K average deal value, 2% branded-search conversion rate, 0.6% direct-session conversion rate. You earned 8 active citations across 5 platforms in a quarter.

Branded-search lift. 4,200 impression delta × 35% CTR = 1,470 sessions × 2% × $30,000 = $882,000. Divided by 8 citations = $110,250 per citation per quarter (gross).

That number is too high. It assumes all of the branded-search lift came from AI exposure. In reality some of it came from other channels — a content piece, a podcast appearance, organic word of mouth.

We use a 30% attribution factor for the branded-search component in B2B SaaS, derived from client interviews and matched-cohort experiments. That brings $110K to $33K per citation gross.

Direct-session assisted conversions. 280 incremental direct sessions × 0.6% × $30,000 = $50,400. Divided by 8 = $6,300 per citation per quarter.

Half-life adjustment. Citations were mostly tactics-category content with measured half-lives between 25 and 65 days. Weighted average around 40 days. Adjustment factor (40/90) = 0.44.

Final. ($33,000 + $6,300) × 0.44 = $17,300 per citation per quarter.

That is on the high end of our portfolio because deal size is large. The $850–$1,400 portfolio median comes out when deal size is smaller and the buying cycle is shorter. The method is the same.

When NOT to model it

Some citations cost more to model than they are worth.

News prompts with sub-7-day half-life. The attribution window does not stabilise before the citation expires. Just count them as visibility wins, not revenue.

Local services. The total quarterly dollar value is usually $200–$500 per citation. The modelling effort is greater than the asset’s value. Track citations as a ranking proxy and skip the dollar model.

Brand-new domains. With under 90 days of history you do not have a control period. Wait one quarter before attempting the model.

What to put in the board deck

Three numbers, not one.

  • Active citations in the quarter (count)
  • Median dollar value per citation (modelled, half-life-adjusted)
  • Total estimated revenue contribution (count × median)

Always include the methodology slide. CFOs respect the half-life adjustment more than the headline number. They have seen too many marketing decks count gross visits as revenue.

If you want the full instrumentation — what to log, how to wire GA4, how to surface this in a dashboard — start with measuring AI citations and the AI pipeline attribution piece. The dollar model is the last step. Most teams skip the first three and end up with a number they cannot defend.